Power_Analysis full narration

Power Analysis : 

Power Analysis A different way of analyzing the data

Power Analysis : 

Power Analysis When to do Do in just about all statistics but makes more sense for more complex studies What is it? The probability of detecting a significant effect when the effect truly exists

Significant, true, huh? : 

Significant, true, huh? So in any experiment you have to rely on many things to determine what is true. We want to see if something is significant What is the likelihood that we will actually be able to see if something is significant That likelihood is power!

What’s power analysis : 

What’s power analysis In stats we look at the probability (p or alpha, same thing) of getting an incorrect positive conclusion There is also a flip side: beta/ Type II error: probability of failing to reject the null hypothesis when it is actually true. It’s significant, but you didn’t detect it.

Refresher on alpha and beta : 

Refresher on alpha and beta

Type I and Type II error : 

Type I and Type II error It’s a trade off As one goes up the other goes down Influenced by 4 things/ parameters

4 parameters to Power analysis : 

4 parameters to Power analysis Beta: Power=1 – beta (probability of rejecting the null hypothesis) Alpha: As alpha increases, power increases. Sample: The sample size. As sample (n) increases, the power increases Reality: The true magnitude of the effect in the population, or degree of departure from the null hypothesis. The larger the magnitude the greater the power.

Figuring power : 

Figuring power Once you know 3 of the factors you can determine the 4th: Like Algebra class- plug n chug So to determine power (1-beta), then you need Alpha Sample size True magnitude of effect in population

Determining true magnitude : 

Determining true magnitude Ok, it’s pretty easy to get the alpha (SPSS gives you that and we do that in class) It’s pretty easy to figure the sample size How the heck do I know what the “true” magnitude is? Can never absolutely know, but can estimate 3 methods of estimation

Power analysis: Determining the size of the pop. effect : 

Power analysis: Determining the size of the pop. effect A review of the relevant literature finds that correlations range from .32 to .43, to be conservative may want to determine if the current study falls in those values- what’s the study’s power to detect a population correlation of .32 (use the lower bound)

Determining size of pop effect: 2nd approach : 

Determining size of pop effect: 2nd approach You might propose an effect that would have some theoretical significance. Might say that a correlation of .5 is necessary or else the study is irrelevant. You would be planning the experiment to detect a correlation of .5

Determining pop. Effect: 3rd approach : 

Determining pop. Effect: 3rd approach Use conventional definitions to determine power such as small, medium or large So for correlations: .15 is small, .30 is a medium correlation, .50 is large for most psychological research What is small, med or large would be influenced by what the literature says

Why do this? : 

Why do this? Can get power analysis, which is informative of the strength of the study. Do this after the study is complete Do prior to study: Can also use to determine what sample size you might need to adequately conduct your experiment. Assume alpha = .05. Then plug n chug the other way to find out necessary minimum sample size. May determine that you don’t have high enough power to run a study so you should Increase sample size (only thing you can really control)

Reading power analysis in articles : 

Reading power analysis in articles Give you the a priori analysis to determine sample size: make assumptions for 3 aspects of the equation and get minimum sample size Post hoc power analysis: Tells you the magnitude/ strength of the study Higher is generally better Want to examine how they determined the population effect

Power analysis on computers : 

Power analysis on computers Option in SPSS for many statistics (All GLM). You can adjust the acceptable significance level in the test to give you different levels of power. How SPSS calculates it: gives observed power when the alternative hypothesis is set based on the observed values. Based on algorithms so beyond that you’ll want to obtain an SPSS book.

Power analysis- online help : 

Power analysis- online help However SPSS is not strongest method. Can only do post-hoc Online help to calculate power: http://www.statsoft.com/textbook/stpowan.html http://www.mp1-pwrc.usgs.gov/powcase/index.html Online calculators http://www.math.uiowa.edu/~rlenth/Power/ A freeware program: G*power (developed in German but translated with a guide) http://www.psycho.uni-duesseldorf.de/aap/projects/gpower/index.html

Power- book help : 

Power- book help Calculating power is different for each type of statistic Book help: Cohen & Cohen textbook for regressions (in e-library), Tabachnick & Fidell textbook a little bit (see index); Statistical Power Analysis for the Behavioral Sciences by Cohen (~$115, library); How Many Subjects? : Statistical Power Analysis in Research by Kraemer (~$35, library); Statistical Power Analysis by Murphy & Myors (e-library)

Gpower program : 

Gpower program